An application of neural nets to the Rome precipitation series (1782–1989)

Abstract

In a recent paper Elsner and Tsonis (1992) have demonstrated the feasibility of short-term prediction of climatic data by means of neural nets. In this paper, the same technique has been applied to the rather long (208 years) Rome precipitation series. At first, some examples of artificially generated time series, ranging from analytical functions to a mathematical model, have been used to test the capability of the method. Then the neural net has been applied to the annual and quarterly amounts of precipitation recorded in Rome. It resulted that this technique, particularly in the case of quarterly values, is capable of giving reasonable short-term predictions.